ThursdAI - The top AI news from the past week

Sally Ann O'Malley, principal engineer at Red Hat who secures and scales containerized AI deployments. Sunil Pai, principal systems engineer at Cloudflare and founder of PartyKit, expert in real-time multiplayer and agent platforms. They talk about agent SDKs and deployment tooling. They discuss container security, agent-to-agent communication, scaling agent infrastructure, and conference trends in AI engineering.

Max Spiro, founder and ML researcher at Pangram Labs who builds AI-detectability tools. He discusses Pangram’s detection research and Chrome extension. Short takes cover false positive targets, detecting AI edits versus full generations, and plans for multimodal detection. The conversation frames detection as a tool for trust, feeds, and cognitive security.

Peter Gostev, AI evaluator and researcher with Arena/LMArena, shares early-access testing impressions of GPT-5.5 and GPT-Image-2. They discuss GPT-5.5’s sustained long-run reasoning and Codex tool integrations. They highlight GPT-Image-2’s 4K outputs, QR and 360 image abilities. Conversation also touches on open-source rises like Kimi K2.6 and Qwen 3.6.

Theodore, product manager at Cognition working on Windsurf 2.0 and agent command centers. Trevor Mons, founding engineer at Marimo building reactive Python notebooks for agent-driven data workflows. Quinn Lahodman-Kramer, co-founder and creator of Gradient-Bang, a voice-driven multiplayer LLM game. They discuss Windsurf 2.0 and Devin integration. Pairing agents with stateful Marimo notebooks. Voice multi-agent game design and stack choices.

VB (OpenAI developer experience lead) and Vincent Koh (OpenClaw maintainer) join Peter Gostev (Arena contributor) and swyx (AI Engineer founder). They discuss Anthropic’s Mythos preview and its security implications. Meta’s Muse Spark and new infra get attention. Z.AI’s GLM 5.1, video model advances, agentic memory tools, and evaluation tooling rounds out the conversation.

Ryan Carson, practitioner who shares agent setups and tooling. Wolfram Ravenwolf, AI evangelist who runs model and agent evaluations. Omar Sanseviero, DeepMind developer-experience rep involved with Gemma releases. They unpack a massive Claude Code source leak, Anthropic’s SessionGate and trust issues, Gemma 4’s licensing and deployment options, and a one-bit quantization breakthrough enabling tiny local models.

Daniel Han, open-source ML engineer and Onslaught contributor who builds tools for fine-tuning and quantization, joins to unpack model compression and deployment. He explains weight quantization, KV cache trade-offs, and real-world limits of TurboQuant. The conversation also reacts to Gemini 3.1 Flash Live, new voice and music models, and key infra and quota dramas in the AI world.

Yam Peleg, an AI practitioner who probes tooling and enterprise implications, and Wolfram Ravenwolf, an engineer who runs WolfBench performance analyses, dig into Anthropic's 1M-context Opus, OpenAI's GPT‑5.4 Mini and Nano, Mistral's new small model, Minimax 2.7's self‑improvement loops, Jensen's OpenClaw remarks, Composer 2 and benchmark showdowns. Short takes on agentic tooling, benchmarks, and model tradeoffs.

Dota, an open-source developer behind Paperclip.ing, explains agent orchestration and heartbeat-driven 'zero-human' workflows. Chris Alexiuk, a NVIDIA product research engineer, outlines NemoTron 3 Super and NVIDIA’s open-source commitments. They discuss AutoResearcher-style autonomous research, the fruit fly brain upload, OpenClaw’s viral adoption, and NVIDIA’s $26B open-source push.

Big model news with a surprise GPT-5.4 drop that adds 1M-context, faster modes, and mid-thought steering. Tense industry politics around Anthropic and a government supply-chain designation. Open-source drama as a prominent researcher departs Qwen and Alibaba responds. New tools and benchmarks like WolfBench and Qwen 3.5 small spark debate about performance and running models locally.